Mean top height vs dominant height: picking the right input for a volume regression

If you're rebuilding a volume regression against remotely sensed canopy height, the first question isn't which algorithm to run. It's which height your old yield tables were actually built on. Mean top height and dominant height get used interchangeably in casual conversation and they are not the same number, and plugging one into an equation calibrated on the other will bias your standing volume estimate in a direction that's hard to catch until a harvest reconciles against it.
What each term actually measures
Mean top height (MTH), the convention most UK and European yield tables use, is the average height of the largest-diameter trees in a stand, typically the 100 largest stems per hectare, following the Hamilton and Christie formulation that's been baked into Forestry Commission yield models for decades. It's a diameter-ranked metric. You identify the biggest trees by dbh, then average their heights.
Dominant height, the term more common in US and some Southern Hemisphere practice, is the average height of trees in the dominant and codominant crown classes. A cruiser ranks trees by crown class to get there, sorting dominant and codominant crowns by eye rather than by diameter. Two cruisers on the same plot can land on slightly different dominant-height figures depending on how they call borderline crowns.
Both exist because Lorey's mean height, the basal-area-weighted average of every tree in a plot, washes out exactly the signal that predicts site productivity. Suppressed and intermediate trees drag the mean down and tell you almost nothing about how fast the site grows dominant stock. Top height and dominant height were built to strip that noise out.
Why this matters for a height-to-volume regression
A canopy surface model built from stereo imagery doesn't see individual stems, crown classes, or dbh rankings. It sees the upper envelope of the canopy, the elevation surface that the matched image points trace across the treetops. What comes out of that surface, whether you sample it as a mean, a 90th-percentile, or a 95th-percentile height per stand polygon, behaves more like top height than like dominant height, because both the CSM and the top-height convention are picking out the upper canopy rather than classifying individual crowns.
That's useful, but it's not a free pass. If the volume table you're regressing against was fit on crown-class dominant height from field cruises decades ago, a flat canopy-surface mean won't match it well, and even a high percentile extracted from the CSM needs to be checked against a handful of field plots before you trust the regression coefficients. The fix most project developers land on is modest: pull several permanent or temporary plots where both the legacy dominant-height figure and the current canopy-surface height are known, regress one against the other, and use that local correction before feeding the stand volume regression input into whatever equation set the holding already runs on.
Building or using a top height curve from imagery
A top height curve, the plot of top height against stand age that underpins most site-index and yield-table work, still needs an age axis you won't get from imagery alone. Stand establishment records or age-class GIS layers fill that gap. What imagery contributes is the height half of the curve, refreshed annually instead of on whatever cycle the last field cruise happened to run on, which matters when a carbon project's reporting cadence doesn't line up neatly with the holding's inventory schedule.
None of this replaces a field cruise entirely. Crown-class calls and dbh rankings for dominant height still require eyes on stems. What a canopy surface model does well is give you a consistent, stand-wide height surface every year, which is exactly the input a volume regression needs once you've pinned down which height definition the rest of your equations expect. Canopy Height Mapping builds that surface from overlapping VHR stereo passes over a defined holding or project area, if you want to see how it lines up against the height metric your current volume tables already use.